CN-121997139-A - Construction method of target domain data set, rolling bearing fault diagnosis method and computer program product

CN121997139ACN 121997139 ACN121997139 ACN 121997139ACN-121997139-A

Abstract

The invention relates to the technical field of rolling bearing fault diagnosis and discloses a construction method of a target domain data set, a rolling bearing fault diagnosis method and a computer program product, wherein the construction method of the target domain data set comprises the steps of selecting an amplitude spectrum of an actual measurement sample as an fitness function of a CS-GWO algorithm, and searching an optimal parameter combination in a VMD algorithm by taking a minimum value of the amplitude spectrum entropy as the fitness function; and calculating the correlation between each modal component and the measured sample, reserving the modal components with high correlation, and carrying out summation reconstruction to obtain target domain data in the target domain data set. According to the invention, the fault characteristics are accurately extracted by using the CS-GWO optimized VMD algorithm, so that the fault characteristic coefficient and the signal to noise ratio are obviously improved, and the noise interference in the actually measured sample is effectively restrained, thereby improving the accuracy and the reliability of the target domain data.

Inventors

WEI CHENGHONG
WEI SHIYONG
GAO XIAOYANG
WU JINGLIANG

Assignees

巨冈精工(广东)股份有限公司

Dates

Publication Date: 20260508
Application Date: 20260128

Claims (10)

1. A method of constructing a target domain dataset, comprising: Collecting vibration signals actually measured by the rolling bearing under four working conditions of normal state, inner ring fault, outer ring fault and rolling body fault as actually measured samples, and constructing an original data set; Selecting an amplitude spectrum of an actually measured sample as an fitness function of a CS-GWO algorithm, and searching an optimal parameter combination in a VMD algorithm by using an extremely small value of an amplitude spectrum entropy as the fitness function Wherein, K is the mode number K, Is a secondary penalty factor; Decomposing the actual measurement sample in the original data set by utilizing the VMD algorithm of the optimal parameter combination to obtain K modal components; And calculating the correlation between each modal component and the actually measured sample, reserving the modal components with high correlation, and summing and reconstructing to obtain the target domain data in the target domain data set.
2. The method for constructing a target domain data set according to claim 1, wherein the VMD algorithm using the optimal parameter combination decomposes the measured samples in the original data set to obtain K modal components, and specifically includes: To actually measure vibration signals Decomposing into K modal components; Is provided with Represents the K modal components obtained after decomposition, Representing the center frequency of each component; for each modal component, hilbert transformation is adopted to calculate an analysis signal, so that a single-side frequency spectrum is obtained: j is the imaginary unit, Is a function of the unit pulse and, Is convolution operation; Modulating the spectrum of the modal function by aliasing of the exponential term of the center frequency corresponding to each modal function: ; Smoothing the frequency spectrum of each modal function by using a Gaussian smoothing method, estimating the bandwidth of the modal function, and solving a constraint variation problem so that the sum of the estimated bandwidths of each modal function is minimum, wherein the constraint variation expression is as follows: ; representing the derivative of the function with respect to time t.
3. The method for constructing a target domain data set according to claim 2, wherein the solving the constraint variation problem is specifically: using a secondary penalty factor Lagrangian multiplier Solving the constraint variation expression, and changing the constraint variation problem into an unconstrained variation problem, namely: ; solving unconstrained variational problem by using alternate direction multiplier method through alternate updating 、 And The saddle point of the augmented lagrangian expression is sought, solving: And (C) sum The updating method of the center frequency comprises the following steps: 。
4. a method of constructing a target domain data set according to claim 3, wherein said calculating the correlation between each modal component and the measured sample, retaining the high correlation modal components and summing the reconstruction, and obtaining the target domain data in the target domain data set comprises: Carrying out correlation analysis on each modal component and the original vibration signal to evaluate the correlation degree, wherein the correlation calculation formula is as follows: ; in the formula, For the signal of the modal component at the sampling point i, As the signal mean value of the modal signal, Is the signal of the original vibration signal at the sampling point i, The signal mean value of the original vibration signal is obtained, and N is the number of sampling points; And preserving modal components with the correlation larger than or equal to a preset correlation threshold value, and summing to reconstruct the original vibration signal.
5. The method as set forth in claim 1, wherein the selecting the magnitude spectrum of the measured sample as the fitness function of the CS-GWO algorithm, and searching for the optimal parameter combination in the VMD algorithm with the minimum value of the magnitude spectrum entropy as the fitness value The method specifically comprises the following steps: initializing parameters of a CS-GWO algorithm and a VMD; The minimum value of the amplitude spectrum entropy is taken as the fitness function, and the CS-GWO algorithm is used for searching the optimal parameter combination in the VMD algorithm Wherein, the calculation formula of the amplitude spectrum entropy is as follows: ; In the above formula, L i is the amplitude spectrum of the modal component u i , H i is the amplitude spectrum entropy of the modal component u i , and N is the length of the modal component.
6. A method of fault diagnosis of a rolling bearing, comprising: based on a nonlinear dynamic equation of the rolling bearing, generating simulation data under four working conditions of normal state, inner ring fault, outer ring fault and rolling body fault, and constructing a source domain data set with a complete fault label; and constructing the target domain data set by adopting the method for constructing the target domain data set according to any one of claims 1 to 5; Randomly dividing a part of the target domain data set and the source domain data set into training sets, dividing the rest part of the target domain data set into test sets, and training a pre-constructed improved alternate migration learning model; And performing fault diagnosis on the full-life rolling bearing by using the trained improved alternate transfer learning model, and outputting a diagnosis result.
7. The method for diagnosing a rolling bearing failure according to claim 6, wherein the construction method of the improved alternate shift learning model is as follows: Constructing a CNN model comprising five convolution layers, two pooling layers and two full connection layers, adding a batch of standardization layers after each convolution layer, adopting a ReLU function as a hidden layer activation function and a Softmax function as an output layer activation function, wherein a first pooling layer is positioned between a second convolution layer and a third convolution layer, and a second pooling layer is positioned before the first full connection layer after the fifth convolution layer; calculating a CORAL loss function after a first convolution layer of the CNN model, and reducing the second-order statistic difference of a source domain data set and a target domain data set; the sum of the MMD penalty function and the classification penalty is calculated at the full connectivity layer, updating the network weights and bias parameters by alternating back propagation.
8. The method for diagnosing a rolling bearing failure in accordance with claim 7, wherein the method for setting the CORAL loss function is as follows: the distance of the second order statistic of the data features in the source domain dataset from the target domain dataset is defined as the CORAL loss function L 1 : , the Frobenius norm of the mean square matrix; Wherein, the For the sample dimension number, the feature covariance matrices of the source domain dataset D S and the target domain dataset D T are C S and C T , respectively; MMD is used to measure the difference in distribution of feature sets D S and D T in the regenerated kernel hilbert space, summing MMD loss with classification loss as a loss function L 2 : , , , for regenerating the nuclear hilbert space; In the above equation, L C is the classification loss between the real tag and the predicted tag, , Mapping function , For the balance coefficient, n S is the number of samples, y is the true label of the sample data, Labels predicted for the classifier.
9. The method for diagnosing a rolling bearing fault according to claim 8, wherein generating simulation data under four working conditions of a normal state, an inner ring fault, an outer ring fault and a rolling body fault based on a nonlinear dynamics equation of the rolling bearing, and constructing a source domain data set with a complete fault label comprises: Correcting a radial displacement excitation function, equivalent contact stiffness and impact force of the rolling body based on Hertz contact theory, and establishing a nonlinear dynamic model of the rolling bearing, wherein the radial displacement excitation function is corrected according to the geometric position of the rolling body in a defect area, and the impact force is corrected according to the fault defect size and the bearing rotating speed; and solving a nonlinear dynamics model of the rolling bearing by a fourth-order Dragon-Gregory tower method, and generating simulation vibration signals under the four working conditions to form a labeled source domain data set.
10. A computer program product comprising computer programs/instructions which, when executed by a processor, implement a method of constructing a target domain dataset according to any of claims 1-5.

Description

Construction method of target domain data set, rolling bearing fault diagnosis method and computer program product Technical Field The invention relates to the technical field of rolling bearing fault diagnosis, in particular to a construction method of a target domain data set, a rolling bearing fault diagnosis method and a computer program product. Background The rolling bearing is used as a core part of the rotary machine, and the running state of the rolling bearing directly influences the reliability and the safety of the equipment, so that the fault diagnosis of the rolling bearing is a key technology for guaranteeing the stable running of a mechanical system. The existing rolling bearing fault diagnosis method is mostly dependent on a large number of laboratory measured data training diagnosis models with labels, but has various pain points in practical application, namely, on one hand, the laboratory measured fault data is high in acquisition cost and long in period, and especially, fault data of bearings with different working conditions and different types are difficult to cover comprehensively, and on the other hand, the simulation mechanism data and the laboratory measured data have domain distribution differences, so that when the model trained by the simulation data is directly applied to the measured data, the diagnosis precision is greatly reduced. The conventional transfer learning method is used for solving the problem of domain distribution difference, but has the defects that a part of the method only adopts a single loss function to measure domain difference, so that characteristic distribution of simulation and measured data is difficult to sufficiently align, effective characteristic extraction and noise reduction pretreatment mechanisms are lacked aiming at noisy measured data, fault characteristics are covered by noise to further reduce diagnosis precision, and meanwhile, under complex scenes such as cross-bearing model, small samples and the like, generalization capability and diagnosis robustness of the conventional method are insufficient, and actual engineering requirements cannot be met. Therefore, in the field, there is a need for a method that can integrate mechanism simulation data and laboratory measured data, and implement fault diagnosis of a high-precision rolling bearing in a complex scene through accurate feature extraction and high-efficiency domain adaptive transfer learning. Disclosure of Invention The object of the present invention is to provide a method for constructing a target domain dataset, a method for diagnosing a rolling bearing failure and a computer program product, which solve or at least partially solve the technical problems mentioned in the background art. To achieve the purpose, the invention adopts the following technical scheme: in a first aspect, the present invention provides a method for constructing a target domain dataset, including: Collecting vibration signals actually measured by the rolling bearing under four working conditions of normal state, inner ring fault, outer ring fault and rolling body fault as actually measured samples, and constructing an original data set; Selecting an amplitude spectrum of an actually measured sample as an fitness function of a CS-GWO algorithm, and searching an optimal parameter combination in a VMD algorithm by using an extremely small value of an amplitude spectrum entropy as the fitness function Wherein, K is the mode number K,Is a secondary penalty factor; Decomposing the actual measurement sample in the original data set by utilizing the VMD algorithm of the optimal parameter combination to obtain K modal components; And calculating the correlation between each modal component and the actually measured sample, reserving the modal components with high correlation, and summing and reconstructing to obtain the target domain data in the target domain data set. Optionally, the decomposing the actually measured samples in the original data set by using the VMD algorithm of the optimal parameter combination to obtain K modal components specifically includes: To actually measure vibration signals Decomposing into K modal components; Is provided with Represents the K modal components obtained after decomposition,Representing the center frequency of each component; for each modal component, hilbert transformation is adopted to calculate an analysis signal, so that a single-side frequency spectrum is obtained: j is the imaginary unit, Is a function of the unit pulse and,Is convolution operation; Modulating the spectrum of the modal function by aliasing of the exponential term of the center frequency corresponding to each modal function: ; Smoothing the frequency spectrum of each modal function by using a Gaussian smoothing method, estimating the bandwidth of the modal function, and solving a constraint variation problem so that the sum of the estimated bandwidths of each modal function is minimum, wherein the con